Determining a reliable strategy for the prediction of ovarian cancer: serial versus parallel testing with a multivariate index assay

Abstract

<h3>Objectives:</h3> The objectives of this study were to evaluate ovarian masses across multiple imaging models in series and in parallel with a second-generation multivariate index assay (MIA2G), and to develop a reliable protocol for ovarian malignancy risk stratification. <h3>Methods:</h3> This is an ongoing randomized controlled trial in Filipino women with ovarian masses in the University of the Philippines - Philippine General Hospital. A comprehensive analysis was performed between various imaging risk classifier methods: IOTA (International Ovarian Tumor Analysis) - Simple Rules, IOTA - LR1, IOTA LR - 2, IOTA-Adnex, Sassone, and Lerner; and a biochemical algorithm: MIA2G (OVERA). The sensitivity, specificity, positive and negative predictive values, and correct classification rate in classifying benign or malignant lesions were computed, as well as their interval estimates. The histopathologic findings served as the ‘gold' standard or variable that represented ‘true' presence of disease. Individual screening performances were assessed using parallel testing between the MIA2G (with conventional cut-off at 5.0) and various imaging risk classifier methods. However, based on previous local data, a higher cut-off at 7.0 was observed such that an additional analysis with the adjusted cut-off was also computed. Similarly, these diagnostic criteria were evaluated in series with the imaging tests followed by the MIA2G at both cut offs (5.0 and 7.0). <h3>Results:</h3> There were 410 tumors identified from 343 women included in the study. Around 20% (67/343) had bilateral masses and only 72% (246/343) of women had histopathologic findings available. The observed study prevalence of women with confirmed malignancy was 31.5% (108/343, 26.6 to 36.7%) and much higher than the expected prevalence of the general population. The MIA2G was noted to be high-risk in 73% (251/343) of all the evaluable subjects using the traditional cut-off at greater than 5.0; but at a higher cut-off of 7.0, this resulted in only 40% (138/343) being classified as high-risk. The individual imaging test sensitivity, specificity, and predictive values calculated showed that the optimal imaging methods were IOTA-LR1, IOTA-LR2 and IOTA-Simple Rules, but the latter had a significant number of inconclusive results. When combined with imaging models in parallel, the conventional MIA2G cut-off (5.0), was noted to perform poorer than the adjusted cut-off (7.0). On the other hand, serial testing showed higher accuracy for imaging tests combined with the conventional MIA2G cut off (5.0). <h3>Conclusions:</h3> In this investigation, ultrasound imaging models and MIA2G are complementary and when used in combination, these provide a better understanding of the malignant potential of an ovarian tumor. MIA2G combined with IOTA-LR2 was the most reliable combination. For parallel testing, an adjusted MIA2G cut-off (7.0) increased overall test accuracy and sensitivity. In serial testing, the conventional MIA2G cut-off (5.0) resulted in overall higher test accuracy and specificity. Applied to our local setting, where the choice of imaging model used for pre-operative evaluation will vary according to the health provider's experience in performing such tests and where there are no standard protocols on the use of biochemical assays, inclusion of such findings in practice guidelines will significantly affect malignancy risk stratification and clinical outcome.